Multiple Imputation using Weighted Quantile Sum Analysis

Consider a set/mixture of continuous, correlated, and censored components/chemicals that are reasonable to combine in an index and share a common outcome. These components are also interval-censored between zero and upper thresholds, or detection limits, that may be different among the components. The `miWQS` package applies the multiple imputation (MI) procedure to the weighted quantile sum regression (WQS) methodology for continuous, binary, or count outcomes. In summary, MI consists of three stages: (1) imputation, (2) analysis, and (3) pooling. First, the missing values are imputed by bootstrapping (Lubin et.al (2004) ), Bayesian imputation, or placing the below the detection limits in the first quantile (BDLQ1) (Ward et.al. (2014) ). Second, the estimate.wqs() function implements WQS regression if the components are complete, imputed, or missing (Carrico et.al. (2014) ) . If the data is missing, BDLQ1 is automatically implemented. Lastly, the pool.mi() function calculates the pooled statistics according to Rubin's rules (Rubin 1987).